I'm going from the explanation of the auction mechanism that's given in the video; correct me if something has changed or I've missed some details.
For people unfamiliar with the proposed auction, it goes something like this: money from the opening bid is burned as dividends. Then the next bid is split into 3 parts:
1) The amount of the previous bid is returned to the previous high bidder.
2) 50% of the remaining amount is also returned to the previous high bidder.
3) The rest is burned as dividends.
I assume that the point of this is that it attracts people who like a good gamble, and that gamblers will create bidding frenzies that raise the price way up. Unfortunately, this auction mechanism has a lot of problems.
I'd like to write this up more formally, but alas, I'm not going to have time to do that for the next couple weeks. So here are a couple casual observations:
1. The only rational opening bid is 0 (or the minimum initial bid). If I bid minimum, then I maximize my potential revenue from the next bidder. As a result, the maximum profit for shareholders is only 50% of the total auction revenue.
2. There is no explicit incentive to place high bids, because at every stage, the lower I bid, either the less I pay for the domain name (because I am the winner) or the more profit I make from the next bidder. The only situation where it makes sense to bid higher than the minimum is when it's a high-profile auction with a low price and I'm afraid that if I only bid minimum, 100 other people also will, so I try to outbid them at this stage. This will make low-profile or end-stage auctions crawl along at a snail's pace.
3. Bidders have an explicit incentive to lie about how much they value the name. There are two kinds of bidders: a) the kind who wants to win, and b) the kind who doesn't care about the name, but just wants to make a quick buck bidding.
4. This mechanism creates all kinds of opportunities to game the system, and this can cost shareholders a lot of money. Example: I value "biophil.p2p" at 1000 DNS, but I'd obviously love to get away with paying less. So I start the auction at 0, and then the very next block, I enter a bid of 1500 DNS from another address. Notice that now, I've only paid 750 (because my first address got half the bid increase), but anybody who wants in now has to pay at least 1500. How is this good for shareholders? I would have been willing to pay 1000, but because the mechanism is so easy to game, I've essentially stolen 250 from the shareholders by a tiny bit of cleverness.
Some of these problems go away if you move to an ebay-style sequential 2nd-price auction, because then everybody's best move (dominant strategy) is to report their value truthfully. 2nd-price auctions have their faults also; if you really want to keep this system of paying bidders back for their bids, would you please consider decreasing the 50% part? You could change this to 10%; then shareholders get 90% of the total revenue, and the gaming-attack I talked about in 4 is much less profitable for the attacker. The attack is still possible, but it doesn't hurt anybody very much. Even 10% is an incentive for speculators and gamblers to come try to create bidding frenzies.
Another idea: condition each bidder's payout by how much larger his bid was than the previous bidder. Example: The bid is at 10. If I bid 11, and then the next guy bids 12, I get 0.5 profit. You could change this to one where if I only bid 11 (which corresponds to a 10% increase), then the most profit I can make is 5% of the next bid increment. But say I bid 15 (a 50% increase): now I'll get 25% of the next bid increment. My numbers don't work (the payout function has to be sublinear), but you get the idea. Now, I'm explicitly incentivized to bid more than the minimum bid at each stage. If you think this is interesting, I'll put a little more thought into it and design a bidder-payout-function that works.
That was a mouthful.
TL;DR: 50% bidder-payout is ripping the shareholders off.